Brand Concept Maps - A Methodology for Identifying Brand Association Networks

Prévia do material em texto

Journal of Marketing Research
Vol. XLIII (November 2006), 549–563549
© 2006, American Marketing Association
ISSN: 0022-2437 (print), 1547-7193 (electronic)
*Deborah Roedder John is Professor and Curtis L. Carlson Chair in
Marketing (e-mail: djohn@csom.umn.edu), and Barbara Loken is Profes-
sor of Marketing (e-mail: bloken@csom.umn.edu), Carlson School of
Management, University of Minnesota. Kyeongheui Kim is Assistant Pro-
fessor of Marketing, University of Toronto (e-mail: kkim@Rotman.
Utoronto.Ca). Alokparna (Sonia) Basu Monga is Assistant Professor of
Marketing, University of Texas at San Antonio (e-mail: alokparna.
monga@utsa.edu). Contributions of the first and second author were equal.
The authors thank Kent Seltman and Lindsay Dingle from the Mayo
Clinic–Rochester for their participation and support. They also thank Lan
Nguyen Chaplin for help with stimuli development and data coding. This
research was sponsored by McKnight grants from the Carlson School of
Management and funding from the Mayo Foundation.
DEBORAH ROEDDER JOHN, BARBARA LOKEN, KYEONGHEUI KIM, and
ALOKPARNA BASU MONGA*
Understanding brand equity involves identifying the network of strong,
favorable, and unique brand associations in memory. This article
introduces a methodology, Brand Concept Maps, for eliciting brand
association networks (maps) from consumers and aggregating individual
maps into a consensus map of the brand. Consensus brand maps
include the core brand associations that define the brand’s image and
show which brand associations are linked directly to the brand, which
associations are linked indirectly to the brand, and which associations
are grouped together. Two studies illustrate the Brand Concept Maps 
methodology and provide evidence of its reliability and validity.
Brand Concept Maps: A Methodology for
Identifying Brand Association Networks
Understanding brand equity involves identifying the net-
work of strong, favorable, and unique brand associations in
consumer memory (Keller 1993). Consumers might associ-
ate a brand with a particular attribute or feature, usage sit-
uation, product spokesperson, or logo. These associations
are typically viewed as being organized in a network in a
manner consistent with associative network models of
memory (see Anderson 1983). This association network
constitutes a brand’s image, identifies the brand’s unique-
ness and value to consumers, and suggests ways that the
brand’s equity can be leveraged in the marketplace (Aaker
1996).
Ideally, firms should be able to measure this network of
brand associations to obtain a brand map, such as the one
for McDonald’s in Figure 1. This map not only identifies
important brand associations but also conveys how these
associations are connected to the brand and to one another.
First, the map pinpoints several associations that are con-
nected directly to the McDonald’s brand, such as “service”
and “value,” and therefore are more closely tied to the
brand’s meaning. Second, the map shows how other associ-
ations are connected to these close brand associations. For
example, “hassle-free,” “convenient,” and “fast” are con-
nected to the “service” association. Third, the map shows
additional linkages between associations. For example, sev-
eral core associations—“meals,” “value,” and “service”—
are connected to one another but are not connected to other
core associations, such as “social involvement.”
However, methodologies for producing brand maps have
been slow to emerge. Many methods are available for elicit-
ing brand associations from consumers, ranging from quali-
tative techniques, such as collages and focus groups, to
quantitative methods, such as attribute rating scales and
brand personality inventories. Techniques such as multi-
dimensional scaling are helpful in understanding how
brands are viewed and what dimensions underlie these per-
ceptions, but these techniques do not identify brand associ-
ation networks—that is, which associations are linked
directly to the brand, which associations are indirectly
linked to the brand through other associations, and which
associations are grouped together.
Two categories of techniques that differ in the way they
derive brand maps are promising in this regard. The first,
which we refer to as “consumer mapping,” elicits brand
maps directly from consumers. Brand associations are
elicited from consumers, who are then asked to construct
networks of these associations as links to the brand and to
one another. Illustrative of this approach is Zaltman’s
Metaphor Elicitation Technique (ZMET), which uses quali-
tative research techniques to identify key brand associations
and then uses in-depth interviews with respondents to
uncover the links between these brand associations (Zalt-
man and Coulter 1995). The second category of techniques,
which we refer to as “analytical mapping,” produces brand
maps using analytical methods. Brand associations are
550 JOURNAL OF MARKETING RESEARCH, NOVEMBER 2006
Figure 1
BRAND MAP FOR MCDONALD’S
Source: Reprinted with permission of The Free Press, a division of Simon & Schuster Adult Publishing Group, from Building Strong Brands, by David A.
Aaker (1996). Copyright by David A. Aaker. All rights reserved.
elicited from consumers, but analytical methods are used to
uncover the network of brand associations. Illustrative of
this approach is network analysis, which uses consumer
perceptions about brands and derives the structure of brand
associations through network algorithms (see Henderson,
Iacobucci, and Calder 1998).
Despite these developments, barriers remain in making
brand-mapping techniques more accessible to marketing
practitioners. In consumer mapping approaches, the process
of eliciting brand maps from individual consumers and
aggregating these individual maps into a consensus brand
map can be labor intensive and require specialized expert-
ise. For example, ZMET requires the use of lengthy per-
sonal interviews conducted by interviewers trained in sev-
eral base disciplines, such as cognitive neuroscience and
psycholinguistics. Analytical mapping techniques offer a
less labor-intensive process for generating maps through the
use of quantitative analyses, but such techniques require
knowledge of statistical techniques that are unfamiliar to
most marketing researchers. For example, network analysis
is a well-known technique in sociology, but it is unfamiliar
to most marketing research firms.
In this article, we offer a new consumer mapping
approach, called Brand Concept Maps (BCM), to answer
the need for a more accessible and standardized method for
producing brand maps. Our approach is easier to administer
than existing consumer mapping techniques, such as
ZMET, and does not require specially trained interviewers
and large time commitments from respondents. In addition,
the BCM offers a flexible approach that is capable of being
used in many research settings, even with large sample sizes
that cover diverse market segments. Compared with exist-
ing analytical mapping techniques, such as network analy-
sis, our approach offers a standardized approach for aggre-
gating individual brand maps using a relatively
straightforward set of rules that do not require knowledge of
specialized statistical techniques.
The remainder of the article proceeds as follows: We
begin by providing more background on consumer mapping
methods and describe ZMET and BCM in detail. Next, we
discuss the first study; we describe the BCM methodology,
illustrate its application, and provide evidence of its reliabil-
ity (split-half reliability) and validity (nomological
validity). We then present a second study that provides evi-
dence of convergent validity, comparing results from the
BCM technique with more conventional ways of measuring
brand perceptions. In the final section, we evaluate the
strengths and weaknesses of the BCM approach as well as
its usefulness for brand management.
CONSUMER MAPPING TECHNIQUES
Consumer mapping techniques can be described in termsof three stages. The first is the elicitation stage, in which
important brand associations are elicited from consumers.
In the second stage, consumers map these elicited associa-
tions to show how they are connected to one another and to
Brand Concept Maps 551
the brand. In the third stage, researchers aggregate these
individual brand maps and associated data to produce a con-
sensus brand map.
In this section, we describe how these stages are accom-
plished in the most well-known consumer mapping tech-
nique, ZMET, and in our technique, BCM. We also evaluate
each technique in terms of criteria that are important across
many branding applications: ease of administration, flexi-
bility across research settings, and quality of the obtained
data in terms of reliability and validity.
ZMET
Description. Zaltman’s Metaphor Elicitation Technique
is designed to “surface the mental models that drive con-
sumer thinking and behavior” (Zaltman and Coulter 1995,
p. 36). It can be used for understanding consumers’
thoughts about brands and product categories (Zaltman and
Coulter 1995).
In the elicitation stage, a small number of participants,
typically 20–25, are recruited and introduced to the topic of
the study (brand). Participants are then given instructions to
take photographs and/or collect a minimum of 12 pictures
of images that will convey their thoughts and feelings about
the topic. Seven to ten days later, participants return with
the requested materials and engage in a two-hour personal
interview to elicit constructs. The personal interview uses
qualitative techniques that tap verbal constructs, such as
Kelly’s repertory grid (respondents identify how any two of
three randomly selected pictures are similar but different
from a third stimulus) and laddering exercises (respondents
specify a means–end chain that consists of attributes, conse-
quences, and values). The interviews also include several
activities aimed at eliciting visual images that represent the
topic of interest. Interviewers are specially trained in these
elicitation techniques and are familiar with base disciplines
(e.g., cognitive neuroscience, psycholinguistics, semiotics)
underlying ZMET.
This is followed by the mapping stage, in which partici-
pants create a map or visual montage using the constructs
that have been elicited. The interviewer reviews all the con-
structs that have been elicited with the respondent and then
asks him or her to create a map that illustrates the important
connections among important constructs.
In the aggregation stage, researchers construct a consen-
sus map that shows the most important constructs and their
relationships across respondents. Interview transcripts,
audiotapes, images, and interviewers’ notes are examined
for the presence of constructs and construct pairs (two con-
structs that are related in some manner). After coding these
data, the researchers make decisions about which constructs
and construct relationships to include in the consensus map
based on how frequently they are mentioned across respon-
dents. The final map contains the chosen elements with
arrows to represent links between constructs.
Evaluation. The primary advantage of ZMET is the thor-
oughness of the procedures for eliciting brand associations;
it uses multiple qualitative research techniques to tap verbal
and nonverbal aspects of consumer thinking. Eliciting brand
associations in this manner is well suited to situations in
which prior branding research is limited or in which deeper
and unconscious aspects of a brand need to be better under-
stood (Christensen and Olson 2002). Reliability and validity
also seem promising. On the basis of validations with sur-
vey data, Zaltman (1997) reports that constructs elicited
using ZMET generalize to larger populations, though the
validity of relationships between constructs (associations)
in consensus maps is still at issue (Zaltman 1997).
The most significant drawbacks of ZMET are related to
accessibility and ease of administration. Accessibility to
practitioners is limited because the procedures for produc-
ing brand maps are not standardized and involve expert
judgment. The technique is also difficult to administer, and
the process is labor intensive (Zaltman 1997). Respondents
must be willing to undergo two interview sessions and
devote additional time to prepare pictures and images for
those interviews. Interviewers with specialized training
determine the composition of the consensus maps through
time-consuming reviews of interview materials. These
requirements limit the flexibility of using ZMET across
research settings, such as focus groups and mall-intercept
studies. In addition, because the elicitation, mapping, and
aggregation stages are so intertwined, ZMET offers little
flexibility for firms with extensive prior brand research that
already know the associations consumers connect to their
brand but want to understand how these associations are
structured in the form of a brand map.
BCM
Background. The BCM methodology is based on a family
of measurement techniques called concept maps. Concept
maps have been used for more than 20 years in the physical
sciences to elicit knowledge people possess about scientific
concepts and how they are interrelated to one another
(Novak and Gowin 1984). Procedures for obtaining concept
maps are flexible, ranging from unstructured methods, in
which respondents generate their own concepts and develop
concept maps with few instructions, to structured methods,
in which lists of concepts are provided and concept map-
ping proceeds with the aid of explicit instructions and con-
cept map examples (for a review, see Ruiz-Primo and
Shavelson 1996). Recently, Joiner (1998) used an unstruc-
tured form of concept mapping to obtain brand maps from
individual consumers. Participants were given a brief set of
instructions, including an example concept map, and were
asked to generate a concept map for a brand by thinking
about the things they associated with the brand and drawing
lines between these associations to show how they were
connected.
However, existing work on concept maps does not offer
procedures for aggregating individual maps into consensus
maps. Individual concept maps obtained using unstructured
methods present many of the same difficulties as those that
ZMET poses. Therefore, procedures for obtaining individ-
ual maps need to be designed with aggregation in mind. To
do so, the BCM incorporates structure into the elicitation,
mapping, and aggregation stages of concept mapping, as we
describe subsequently.
Description. The BCM method provides a map showing
the network of salient brand associations that underlie con-
sumer perceptions of brands. In the elicitation stage,
researchers identify salient associations for the brand.
Existing consumer research can be used for this purpose, or
552 JOURNAL OF MARKETING RESEARCH, NOVEMBER 2006
1The BCM elicitation procedure differs from standard elicitation proce-
dures in attitude research in at least two respects. First, the open-ended
elicitation questions may differ somewhat from standardized elicitation
questions about favorable and unfavorable attributes (or consequences)
used in some attitude research (Fishbein and Ajzen 1975). Second, the
number of associations used for the BCM procedure is typically larger than
the ±7 rule used in some attitude research (Fishbein and Ajzen 1975).
a brief survey can provide the necessary information. The
process for identifying salient associations should conform
to four criteria, guided by procedures for obtaining salient
beliefs in attitude research (e.g., Fishbein and Ajzen 1975).1
First, data used to identify salient associations should be
gathered from the same consumer population as the one
being used in the mapping stage. Second, data used to iden-
tify salient associations should be based on consumer
responses to open-ended questions (e.g., “When you think
of [brand], what comes to mind?”). Open-ended questions
allow consumersto voice whatever brand associations are
most accessible and important to them in their own words.
Third, the most frequently mentioned brand associations
should be selected to form the final set. For our procedure,
we include brand associations that at least 50% of respon-
dents mentioned. Fourth, in selecting the exact phrasing for
salient brand associations, it is important to retain wording
that the consumers use rather than wording that researchers
or managers more commonly use.
To begin the mapping stage, respondents are asked to
think about what they associate with the brand. Salient
brand associations (selected from the first stage) mounted
onto cards are shown to respondents to aid in this process.
Respondents are asked to select any of the premade cards
that reflect their feelings about the brand. As a check to
ensure that all salient brand associations have been included
on the cards, blank cards are made available for respondents
who want to add additional associations to the set. Then,
respondents are shown an example of a BCM and are given
instructions on building their own brand map. Respondents
use the brand associations they have selected and connect
them to one another and to the brand, using another set of
cards with different types of lines (single, double, or triple)
to signify the strength of the connection between
associations.
In the aggregation stage, individual brand maps are com-
bined on the basis of a set of rules to obtain a consensus
map for the brand. As we describe subsequently, these rules
require no specialized knowledge of quantitative or qualita-
tive research methods. Frequencies are used to construct a
consensus map, showing the most salient brand associations
and their interconnections.
Evaluation. The BCM method incorporates structure into
the elicitation, mapping, and aggregation stages to provide a
technique that is easier to administer and analyze. Inter-
viewers need minimal training, and respondents can com-
plete the mapping procedure in a relatively short time (15–
20 minutes). The BCM method also provides flexibility.
Prior consumer research can often be used in the elicitation
stage, enabling researchers to proceed with the mapping
and aggregation stages without further time and expense.
Respondents can complete brand maps relatively quickly,
making the technique suitable for many data collection set-
tings and affording the opportunity to collect larger samples
than ZMET. This, along with more standardized aggrega-
tion procedures, enables firms to collect brand maps for dif-
ferent market segments or geographic areas.
However, the BCM has drawbacks as well. In most cases,
the BCM reveals accessible brand associations and connec-
tions. However, associations that require more in-depth
probing are unlikely to surface with this technique. Most of
the representations are verbal in nature as well. Further-
more, the reliability and validity of consensus brand maps
using BCM requires examination. Although individual con-
cept maps may be valid, consensus maps pose additional
challenges, particularly with regard to aggregation bias that
can adversely affect reliability and validity.
We address these issues in Study 1. We illustrate the use
of the BCM in a real branding context and provide addi-
tional details about the elicitation, mapping, and aggrega-
tion procedures. We also evaluate reliability and validity for
the BCM methodology.
STUDY 1
In this study, we illustrate the use of the BCM in the con-
text of a premier health care brand, the Mayo Clinic. This
afforded us several opportunities to test the capabilities of
the BCM technique. First, the Mayo Clinic is a complex
brand with many salient brand associations, such as “leader
in medical research,” “best doctors in the world,” and
“known worldwide.” This complexity provided a strong test
of the BCM because large numbers of brand associations
can be combined in almost infinite ways in a network struc-
ture, making it difficult to obtain a consensus brand map.
Second, the Mayo Clinic brand elicits a wide variety of
associations, including attributes (e.g., “best doctors in the
world”), personality traits (e.g., “caring and compassion-
ate”), and emotions (e.g., “it comforts me knowing that
Mayo Clinic exists”). This provided an opportunity to test
whether the BCM would be able to incorporate different
types of associations into consensus brand maps. Finally,
the Mayo Clinic is a brand with distinct user segments
(patients versus nonpatients), which enabled us to test
whether BCM would work equally well for users (who
share experiences and similar brand associations) and
nonusers (who are more heterogeneous and likely to have
fewer brand associations in common).
Elicitation Stage
To begin, we selected a set of salient brand associations
for the Mayo Clinic. First, we examined prior consumer
research conducted by the Mayo Clinic, focusing our atten-
tion on responses to open-ended questions about the brand.
We developed frequency counts of how often certain brand
associations were mentioned, and we selected those that at
least 50% of the respondents mentioned. We submitted
these selections for review to the Mayo Clinic brand team,
who added a few more associations of particular interest to
them. We also consulted with members of the brand team to
finalize the exact wording of the brand associations. The
result was a set of 25 brand associations to be used in the
mapping stage.
Mapping Stage
Sample. A total of 165 consumers from two midwestern
cities participated in the study. Ninety participants were
current or former patients at the Mayo Clinic. Patients were
randomly selected from the Mayo Clinic database, sent a
Brand Concept Maps 553
prenotification letter from the Mayo Clinic asking for their
participation, and then recruited by telephone by marketing
research firms in both cities. Seventy-five participants were
nonpatients who were recruited and screened by marketing
research firms. Age and gender quotas were used for both
samples to obtain a broader set of respondents. All partici-
pants received monetary compensation for their
participation.
Procedure. Marketing research firms in both cities con-
ducted one-on-one interviews. Respondents were told that
they were participating in a consumer study of health care
organizations and had been chosen to answer questions
about the Mayo Clinic. Respondents were encouraged to
express their own opinions, whether positive or negative,
and were told that the researchers were not employees of
the Mayo Clinic.
Participants were guided in building their brand maps in
four steps. First, participants were asked to think about the
following question: “What comes to mind when you think
about the Mayo Clinic?” To help them with this task,
respondents were shown a poster board that contained 25
laminated cards, with a different brand association for the
Mayo Clinic printed on each card. Respondents were told
that they could use any of the cards on the poster board and
could add additional thoughts or feelings by writing them
down on blank laminated cards provided. All the chosen
cards were put onto a second poster board to complete this
step.
The second step involved explaining the nature and pur-
pose of the BCM. Respondents were shown a BCM of the
Volkswagen Beetle (see Figure 2). This example was used
to describe the types of associations that might be included
on the map, how associations might be linked to the brand
(directly linked, such as “inexpensive to drive,” or indirectly
linked, such as “good mpg [miles per gallon]”), and how
associations might be linked to one another (e.g., “good
mpg” causes a Volkswagen to be “inexpensive to drive”).
The Volkswagen Beetle map also included different types of
lines that connected associations—specifically, single, dou-
ble, or triple lines. Participants were told that these lines
indicated how strongly an association was connected to the
brand or to another association, with more lines indicatinga
stronger connection.
Third, respondents developed their brand map for the
Mayo Clinic. Participants were given a blank poster board,
with the brand (Mayo Clinic) in the center. They were
instructed to use the laminated cards they had previously
selected and were given different types of lines (single, dou-
ble, or triple) for connecting the laminated cards on their
poster board. Respondents had as much time as they needed
and were allowed to look at the Volkswagen Beetle example
for reference.
In the fourth step, participants were asked to indicate
their feelings about the brand using a number between 1
(“extremely negative”) and 10 (“extremely positive”),
which was then marked on the brand map next to the Mayo
Clinic name. Participants completed several questions about
prior experience and familiarity with the Mayo Clinic as
well as basic demographics. Respondents were then
thanked, debriefed about the study, and dismissed. On aver-
age, respondents completed the entire brand concept map-
ping procedure in 15–25 minutes.
Aggregation Stage
Measures. We first coded information from each respon-
dent’s map in terms of (1) the presence of each of the 25
brand associations, (2) the type of line (single, double, or
triple) connecting each association to the brand or to
another association, (3) the level at which each association
was placed on the map (e.g., Level 1 = connected to brand,
Level 2 = connected under a Level 1 association), and (4)
which brand associations were linked above and below each
association on the map. At this point, we also analyzed
brand associations that the respondents added during the
mapping procedure to determine whether any occurred fre-
quently enough to be added to the original set. None were
mentioned by more than 4% of respondents, so we excluded
them from further analysis. However, we maintained a list
of added associations in case they represented emerging
perceptions of the brand that deserve further management
attention.
We aggregated the coded data to obtain several measures
for constructing the consensus brand map. Measures for the
patient sample appear in Table 1. “Frequency of mention” is
the number of times that a brand association occurs across
maps. In Table 1, “expert in treating serious illnesses” was
the most frequently mentioned association. “Number of
interconnections” represents the number of times that a
brand association is connected to other brand associations.
The belief and attitude structure literature often views inter-
connectivity as indicative of how “central” an element is
within an overall belief system (Eagly and Chaiken 1993;
Rokeach 1968). In Table 1, “expert in treating serious ill-
nesses” had the most interconnections to other brand asso-
ciations. Frequently mentioned associations with many
interconnections are the strongest candidates for being cho-
sen as “core” brand associations on the consensus brand
map.
The next three measures in Table 1 indicate where core
brand associations should be placed on the consensus brand
map, linked directly or indirectly to the brand. “Frequency
Figure 2
BCM EXAMPLE
554 JOURNAL OF MARKETING RESEARCH, NOVEMBER 2006
Core Associations First-Order Associations
Brand Associations
Frequency 
of Mention
Number 
of Inter-
connections
Frequency of
First-Order
Mention
Ratio of
First-Order
Mention (%)
Subordinate
Connections
Super-
ordinate
Connections
Expert in treating serious illnesses 64 75 34 53.1 30 45
Latest medical equipment and technology 60 62 22 36.7 38 24
Leader in medical research 54 60 41 75.9 13 44
Known worldwide 54 57 37 68.5 17 27
Top-notch surgery and treatment 53 44 21 39.6 32 22
Best doctors in the world 51 54 29 56.9 22 52
World leader in new medical treatments 51 74 23 45.1 28 41
Can be trusted to do what’s right for patients 51 69 22 43.1 29 25
Doctors work as a team 50 54 20 40.0 30 34
Best patient care available 49 64 33 67.3 16 45
Treats patients with rare and complex illnesses 49 61 23 46.9 26 18
Can figure out what’s wrong when other doctors can’t 49 44 15 30.6 35 22
Publishes health information to help you stay well 44 57 19 43.2 25 9
Approachable, friendly doctors 44 34 15 34.1 29 2
Caring and compassionate 42 50 19 45.2 23 19
Treats famous people from around the world 38 42 13 34.2 25 0
It comforts me knowing Mayo exists if I ever need it 36 25 19 52.8 18 15
People I know recommend Mayo 30 33 19 63.3 11 4
Leader in cancer research and treatment 29 15 11 37.9 18 5
Cares more about people than money 27 23 14 51.9 13 7
Court of last resort 12 20 5 41.7 7 1
Hard to get into unless very sick or famous 5 8 1 20.0 4 1
Very big and intimidating 3 5 3 100.0 0 4
Expensive 3 4 1 33.3 2 1
Uses its reputation to make money 3 3 1 .0 2 1
Notes: N = 90 respondents. Core brand associations are in bold, and first-order brand associations are in bold italics.
Table 1
STUDY 1: BCM MEASURES FOR PATIENTS
of first-order mentions” is a count of the number of times
that a brand association is directly linked to the brand
across maps. In Table 1, “leader in medical research” was
the association most frequently connected in a direct way to
the Mayo Clinic brand. “Ratio of first-order mentions” is
the percentage of times that a brand association is linked
directly to the brand when it is included on a brand map.
According to Table 1, 75.9% of patients who included
“leader in medical research” on their brand maps placed
this association as a direct link to the Mayo Clinic brand.
“Type of interconnections” indicates how frequently a
brand association is placed above other associations (super-
ordinate) or below other associations (subordinate) across
maps. As Table 1 shows, patients frequently mentioned “lat-
est medical equipment and technology” but placed it more
in a subordinate position (38 maps) than in a superordinate
position (24 maps). Associations linked directly to the
brand on a frequent basis with more superordinate than sub-
ordinate connections are strong candidates for being
directly connected to the brand in the consensus brand map.
Procedure. We used a five-step process to develop a con-
sensus brand map for Mayo Clinic patients and nonpatients
(see Table 2). In the first step, we identified the core brand
associations that would be placed on the map. We used two
measures for this purpose: frequency of mention and num-
ber of interconnections. We identified associations that were
included on at least 50% of the maps as core brand associa-
tions, consistent with cutoff levels in content analyses of
brand/product attributes, beliefs, and values (Reynolds and
Gutman 1988; Sirsi, Ward, and Reingen 1996; Zaltman and
Coulter 1995). We also included associations with border-
line frequencies (45%–49%) if the number of interconnec-
tions was equal to or higher than that of other core brand
associations, consistent with the idea that high interconnec-
tivity signals the centrality of associations or beliefs. Apply-
ing these rules, we found 12 core brand associations for
Mayo Clinic patients (see Table 1).
In the second step, we began the process of building the
consensus map by identifying which core brand associa-
tions should be linked directly to the Mayo brand. We iden-
tified these core brand associations (first-order associations)
using three measures: frequency of first-order mentions,
ratio of first-order mentions, and type of interconnections.
We selected associations with ratios of first-order mentions
to total mentions of at least 50%, with more superordinate
than subordinate connections, as first-order associations.
Applying these rules to the patient data in Table 1, we
selected six core brand associations as first-order associa-
tions, which appear as direct links to the Mayo Clinic brand
in the consensus brand map (see Figure 3).
In the third step, we placed the remaining core brand
associations on the map. They needed to be linked to at
least one of the first-order brand associations; important
links between the 12 core brand associations also neededto
be placed on the consensus map. To do so, we first counted
how frequently links between specific associations occurred
across maps. We then compiled a frequency count of how
many different association links were noted on one map,
two maps, three maps, and so on. As we show in Figure 4,
109 different association links appeared on only one patient
map, 42 different association links appeared on two patient
maps, 24 different association links appeared on three
patient maps, and so on. These frequencies represent links
between associations in one direction only; the vast major-
ity of possible association links (394 of a possible 600)
never occurred on a single map.
Brand Concept Maps 555
Step Measures Rules
1. Select core brand associations Frequency of mention
Number of interconnections
Select brand associations that are
•Included on at least 50% of maps.
•Included on 45%–49% of maps if the number of connections the number
of connections for core associations we identified previously.
2. Select first-order brand associations Frequency of first-order mentions
Ratio of first-order mentions
Type of interconnections
Select core brand associations that
•Have a ratio of first-order mentions to total mentions of at least 50%.
•Have more superordinate than subordinate interconnections.
3. Select core brand association links Frequencies for association links Select core brand association links by
•Finding inflection point on frequency plot.
•Inflection point = target number.
•Including all association links that appear on or above the target number
of maps.
4. Select non–core brand association 
links
Frequencies for association links Select non–core brand association links that are
•Linked to a core brand association.
•Linked on or above the target number of maps.
5. Select number of connecting lines Mean number of lines used per 
link
Select single, double, or triple lines for each brand association link by
•Determining the mean number of lines used per link.
•Rounding up or down to the next integer number (e.g., 2.3 = 2).
Table 2
AGGREGATION RULES FOR BCM
We used these frequencies to select which association
links would be included in the consensus map, looking for a
sharp increase in frequency counts on the graphs (inflection
point). In Figure 4, the inflection point occurs at five; the
decision rule was to include all core association links found
on at least five maps in the consensus brand map. Twenty-
two links met the criteria, but only 12 of these were links
between core brand associations; the remaining links were
between core and non–core associations or between two
non–core associations. We placed the 12 links between core
brand associations on the consensus map to complete this
step.
In the fourth step, we added important links between core
and non–core brand associations to the consensus map. As
we noted previously, several of the frequently mentioned
links were between core and non–core brand associations.
Although the consensus brand map could be restricted to
core brand associations, it is often important for managers
to see associations that drive consumer perceptions of the
core brand associations. We added these links to the consen-
sus map; we represented the non–core brand associations
with dotted lines to distinguish them from the more impor-
tant core brand associations.
In the fifth step, we placed lines (single, double, or triple)
on the map to signify the intensity of the connection
between associations. For each association link, we com-
puted the mean number of lines respondents used and
rounded up or down to the nearest integer (e.g., 2.3 = 2) to
determine how many lines to use on the consensus brand
map. For example, in the patient map, we decided to use a
double line between “best patient care available” and “can
be trusted to do what’s right for patients” on the basis of the
mean value of the number of lines (M = 2.1) that patients
used to connect these two associations on their maps (see
Figure 3).
Consensus maps. The consensus brand maps for patients
and nonpatients appear in Figure 3. As we expected,
patients had consensus maps with more core brand associa-
tions, more first-order associations, more association links,
and stronger connections between associations. Patients
also included brand associations such as “caring and com-
passionate” and “cares more about people than money,”
which capture patient experiences. However, many core
brand associations appeared across both patient and non-
patient maps. Associations such as “leader in medical
research” and “known worldwide” are accessible to both
groups through Mayo Clinic press releases, medical
newsletters, and word of mouth.
How well do these consensus maps summarize the brand
perceptions of patients and nonpatients? As a check on our
aggregation procedures, we compared individual brand
maps with consensus brand maps for patients and non-
patients in two ways. First, following a procedure used for
ZMET, we determined the number of individual maps,
selected at random, that was needed to capture at least 70%
of all core brand association links found in the consensus
maps for patients and nonpatients (see Zaltman and Coulter
1995). The logic here is that a small number of individual
maps should be able to reproduce the association links in
the consensus map if the aggregation procedure has been
successful. In our case, it took 12 patient maps to reproduce
at least 70% of the core brand association links found in the
patient consensus map, and 7 nonpatient maps were needed
to reproduce at least 70% of the core brand association links
found in the nonpatient consensus map. Note that these
numbers represent relatively small samples of individual
maps from patients (13% of maps) and nonpatients (9% of
maps).
Second, we compared individual with consensus brand
maps to determine how well the consensus maps captured
the core brand associations found in individual brand maps.
For example, if an individual’s map includes 12 brand asso-
ciations, how many of these are core brand associations
found on the consensus map? For patients (nonpatients), we
556 JOURNAL OF MARKETING RESEARCH, NOVEMBER 2006
Figure 3
STUDY 1: CONSENSUS BCM FOR MAYO CLINIC
A. Patients
Notes: N = 90 patients, and N = 75 nonpatients. The solid-line circle signifies core associations, and the dashed-line circle signifies non–core associations.
B. Nonpatients
Brand Concept Maps 557
Figure 4
STUDY 1: ANALYSIS OF BRAND ASSOCIATION LINKS FOR
PATIENTS
found that 66% (65%) of the brand associations shown on
the individual maps were pictured as core brand associa-
tions on the consensus map. Furthermore, we checked on
the intensity of the association links by weighting each
brand association shown on an individual map by the num-
ber of lines (single, double, or triple) and attaching a
valance to this number (+ = positive association; – = nega-
tive association). We then divided this number by a similar
one that we computed for the core brand associations shown
on the consensus maps. We found percentages similar to
those for the unweighted analysis: 68% for patients and
68% for nonpatients. Taken together, these analyses indi-
cate that consensus maps capture approximately two-thirds
of the content of individual brand maps, which appears
more than reasonable given the inherent heterogeneity of
individual brand perceptions.
Reliability and Validity Analyses
The BCM method is able to capture the network of brand
associations underlying consumer perceptions of a brand, as
illustrated by the Mayo Clinic application, but does the
BCM satisfy standard measurement criteria, such as relia-
bility and validity? We pursued an answer to this question
using traditional methods of measure validation (see
Churchill 1979). We assessed split-half reliability to deter-
mine how consistent the obtained consensus brand maps
would be across multiple administrations of the technique.
We examined nomological validity by comparing consensus
brandmaps from known groups (expert versus novice con-
sumers) to determine whether the maps reflect expected
expert–novice differences. If so, these results would add to
our confidence that the BCM measures what it purports to
measure.
Split-half reliability. Using the patient sample, which we
chose for its larger sample size, we randomly divided the
individual concept maps into two halves. For each half, we
aggregated individual brand maps into a consensus map. A
comparison of the maps (see Figure 5) suggests a reason-
able degree of consistency. Each map has 17 brand associa-
tions, with 16 associations shared across maps. The first
map has 5 first-order associations, all connected to the
2Each core belief can be linked to any of the other 11 core beliefs or to
the Mayo Clinic brand. For example, possible links for Core Belief 1 are
1–2, 1–3, 1–4, 1–5, 1–6, 1–7, 1–8, 1–9, 1–10, 1–11, 1–12, and 1–Mayo;
additional possible links for Core Belief 2 are 2–3, 2–4, 2–5, 2–6, 2–7,
2–8, 2–9, 2–10, 2–11, 2–12, and 2–Mayo. Counting the number of non-
duplicated links in this way results in 78 links.
Mayo Clinic brand with triple lines, except for a two-line
connection with “known worldwide.” The second map fea-
tures the same first-order associations, connected by the
same number of lines, though there is one additional associ-
ation (“world leader in new medical treatments”). Many of
the links between associations are the same as well.
To obtain quantitative measures of split-half reliability,
we coded each split-half map for the presence or absence of
(1) each of the 25 brand associations as a core association,
(2) each of the 25 brand associations as a first-order associ-
ation, and (3) each of the 300 possible links among the 25
brand associations. We coded presence of a brand associa-
tion or association link as 1, and 0 otherwise. We then com-
puted correlations across split-half maps, which were high-
est for the presence of core brand associations (φ = .92, p <
.01; N = 25), moderately high for the presence of first-order
brand associations (φ = .78, p < .01; N = 25), and moderate
for the presence of specific brand association links (φ = .50,
p < .01; N = 300). Overall, the split-half reliability levels
appear acceptable, even though the reliability of specific
association links is considerably lower because of the sheer
number of possible links and the conservative nature of the
test, which credits only direct links between associations.
For example, the “best doctors in the world” → “can figure
out what’s wrong when other doctors can’t” link is coded as
being present in Half 2 (Figure 5, Panel B) but not in Half 1
(Figure 5, Panel A), even though Half 1 contains the link
embedded within a chain of associations (“best doctors in
the world” → ”doctors work as a team” → ”can figure out
what’s wrong when other doctors can’t”).
We conducted a second analysis to provide further data
about the reliability of the brand association links shown in
the consensus map. We coded each split-half map and the
patient consensus map for the presence or absence of each
of the 78 possible links between the 12 core beliefs and the
Mayo Clinic brand.2 We coded presence of a brand associa-
tion link as 1, and 0 otherwise. We then computed correla-
tions for each split-half map with the consensus map, show-
ing a moderately high degree of reliability for the first split
half (φ = .75, p < .01; N = 78) and the second split half (φ =
.78, p < .01; N = 78). The correlation between split halves
was moderate as before (φ = .54, p < .01; N = 78). A similar
analysis examining the strength of the association links
(single, double, or triple lines) between all 78 possible links
indicated even higher correlations between the consensus
map and Half 1 (r = .75, p < .01; N = 78), the consensus
map and Half 2 (r = .84, p < .01; N = 78), and both split
halves (r = .64, p < .01; N = 78). Using the correlations
between each split half and the consensus map as an indica-
tor of reliability, we obtained a coefficient alpha of .70 for
the presence of association links and .78 for the strength of
association links, both meeting acceptable levels of
reliability.
Nomological validity. We used a known-groups approach
for assessing nomological validity, comparing consensus
558 JOURNAL OF MARKETING RESEARCH, NOVEMBER 2006
Figure 5
STUDY 1: CONSENSUS BCM FOR SPLIT HALVES
A. Half 1
Notes: N = 45 patients per each half. The solid-line circle signifies core associations, and the dashed-line circle signifies non–core associations.
B. Half 2
Brand Concept Maps 559
brand maps for respondents who differed in familiarity with
the Mayo Clinic. Because familiarity is a dimension of
expertise, we expected to find several expert–novice differ-
ences in our comparisons. Experts typically have knowl-
edge structures that are more complex and highly inte-
grated, which would translate into more brand associations,
more brand association links, stronger brand association
links (e.g., more double and triple lines), and greater hierar-
chical structuring (e.g., more third- or fourth-order associa-
tions) in a consensus map (see Novak and Gowin 1984).
Because familiarity can breed stronger feelings and emo-
tions, we also expected experts to have more brand associa-
tions with relationship connotations, such as “caring and
compassionate” and “can be trusted to do what’s right for
patients.”
We divided respondents into two groups: very familiar
and somewhat familiar. As we expected, the vast majority of
patients (81%) were very familiar, but a substantial percent-
age of nonpatients (21%) also considered themselves very
familiar. Many nonpatients knew someone who had been
treated at the Mayo Clinic and could possibly have been
involved in their treatment. The majority of nonpatients
(56%) and a sizable number of patients (17%) identified
themselves as being somewhat familiar. To obtain reason-
able sample sizes for analysis, we limited our analysis to
the “very familiar” and “somewhat familiar” groups.
To assess whether the BCM was capable of picking up
expert–novice differences, we conducted two types of
analysis. First, we used our aggregation procedures to pro-
duce a consensus brand map for both familiarity groups
(see Figure 6). A comparison of these maps shows that the
map for the very familiar group has a more complex struc-
ture, with more brand associations and more interconnec-
tions between associations.
We performed a second analysis to determine whether
these findings could be corroborated with the BCM at the
individual level. This also provided a check on our aggrega-
tion procedures, evaluating whether expert–novice differ-
ences found in the composite brand maps were reflective of
expert–novice differences in individual brand maps. For this
analysis, we coded each respondent’s brand map for the fol-
lowing features: (1) number of brand associations; (2) num-
ber of brand associations at the first, second, third, and
fourth+ levels; (3) number of relationship brand associa-
tions; (4) number of links between brand associations; and
(5) number of single, double, and triple lines. Measures
similar to these have been used in the concept mapping lit-
erature to evaluate the structural complexity of knowledge
structures (see Novak and Gowin 1984) and to examine dif-
ferences between groups that vary in expertise, instruction,
or performance (see, e.g., Joiner 1998; Wallace and Mintzes
1990).
Means and standard deviations for both familiarity
groups appear in Table 3. An analysis of variance revealed
that the very familiar group had brand maps with more
brand associations, more relationship associations, more
brand association links, stronger brand association links (a
greater number of triple lines), and more hierarchical
branching (more third-level links). Thus, the expert–novice
findings from this analysis converge with those we obtained
using the consensus brand maps. The expected expert–
novice differences emerge clearly, providing evidenceof
nomological validity and evidence that the consensus brand
maps capture the essence of individual maps without
noticeable aggregation bias.
Discussion
In this study, we illustrated the use of the BCM in an
actual branding application. We also obtained evidence of
reliability and validity, increasing our confidence that the
BCM yields consensus brand maps that are valid depictions
of the consumer brand perceptions.
An important question at this point is whether the BCM
has predictive validity as well. Do individual brand maps
predict a consumer’s attitude toward the brand? Do features
of the consensus brand maps predict overall attitudes
toward the brand? Recall that our mapping procedure
includes a ten-point attitude scale that can be used for tests
of predictive validity. In our case, attitudes toward the Mayo
Clinic were extremely positive across participants, hamper-
ing our ability to perform a full range of predictive validity
analyses. However, we were able to demonstrate the predic-
tive validity of individual brand maps through a simple
cluster analysis. Using cluster analysis, we identified two
groups of individuals with similar brand associations on
their maps (ncluster1 = 97, ncluster2 = 68). Because these clus-
ters view the brand in different ways, their brand attitudes
should vary as well. Indeed, in comparing clusters on atti-
tudes toward the Mayo Clinic, we found significant differ-
ences in attitudes (Mcluster1 = 8.90, Mcluster2 = 9.69; t(1,
163) = 13.63, p < .01).
Another question that can be raised is whether the BCM
produces data that are consistent with more established
research methodologies. Do features of the individual brand
maps correlate well with results from standard survey
research techniques? In the next study, we pursue evidence
along these lines by assessing convergent validity. We com-
pare consumer perceptions of the Mayo Clinic brand using
the BCM and traditional attribute rating scales. Although
the BCM is designed to capture the network of brand asso-
ciations, which is beyond the purpose of attribute rating
scales, there should nevertheless be some convergence
between them. For example, if consumers agree strongly
with the statement that the Mayo Clinic has the “best doc-
tors in the world,” this association should emerge as a core
brand association in brand maps produced using the BCM.
STUDY 2
Method
Sample. A new sample of respondents was recruited for a
mall-intercept study. Shoppers between the ages of 21 and
75 with at least a high school education, at least some
familiarity with the Mayo Clinic, and no employment his-
tory with the Mayo Clinic were invited to participate for a
$3 incentive. Quotas for age groups and gender were estab-
lished to obtain a broader sample. Twenty-nine participants
were asked about their perceptions of the Mayo Clinic
using the BCM (BCM condition), and 20 participants pro-
vided their perceptions of the Mayo Clinic by answering a
battery of attribute rating scales (attribute-rating-scales
condition).
Procedure. Participants were randomly assigned to one
of the procedure conditions and were interviewed individu-
560 JOURNAL OF MARKETING RESEARCH, NOVEMBER 2006
Figure 6
STUDY 1: CONSENSUS BCM FOR FAMILIARITY GROUPS
A. Very Familiar
Notes: N = 88 for very familiar group, and N = 57 for somewhat familiar group. The solid-line circle signifies core associations, and the dashed-line circle
signifies non–core associations.
B. Somewhat Familiar
Brand Concept Maps 561
Very
Familiar 
Somewhat
Familiar
Total number of associations 12.01a 10.04b
(4.44) (3.94)
Total number of links 12.03a 10.04b
(4.46) (3.94)
Number of first links 5.35 4.79
(3.17) (3.05)
Number of second-level links 4.38 3.75
(2.73) (2.75)
Number of third-level links 1.69a 1.11b
(1.19) (1.20)
Number of fourth-level (or higher) links .59
(1.02)
.37
(.98)
Number of relationship association links 2.34a
(1.70)
1.30b
(1.16)
Number of first-order relationship
association links
.92a
(1.12)
.51b
(.83)
Number of single lines 2.68 2.72
(2.14) (2.59)
Number of double lines 4.06 3.94
(2.45) (2.00)
Number of triple lines 5.27a 3.35b
(2.90) (3.02)
Notes: N = 88 for very familiar group, and N = 57 for somewhat famil-
iar group. Cells with different superscripts differ from each other at p <
.05. Standard deviations are in parentheses.
Table 3
STUDY 1: COMPARISON OF BCM FOR FAMILIARITY GROUPS
Familiarity
ally by an employee of a mall-intercept research firm.
Respondents were told that they were participating in a con-
sumer study of health care organizations and would be
answering questions about the Mayo Clinic. Participants
were encouraged to express their opinions, whether positive
or negative, and were also told that the researchers were not
employees of the Mayo Clinic.
Respondents in the BCM condition constructed a brand
map using the same procedure described in Study 1. How-
ever, we modified the set of brand associations in several
ways. First, we included several foils, consisting of positive
statements that are not usually associated with the Mayo
Clinic, such as “has well-regarded drug and alcohol rehab
services” and “has many convenient locations.” We
included these to assess whether the mapping procedure,
which provides respondents with a prespecified set of brand
associations, biases consumers toward including more posi-
tive associations than those needed to reflect their view of
the brand. Second, we included more negative brand associ-
ations, such as “big and impersonal” and “only for the rich
and famous,” to encourage consumers to select negative
associations during the mapping stage if they had negative
perceptions of the brand.
Participants in the attribute-rating-scales condition com-
pleted a survey about the Mayo Clinic. The survey con-
tained 23 questions about the Mayo Clinic, such as “Do you
agree or disagree that the Mayo Clinic has excellent doc-
tors?” and “Do you agree or disagree that the Mayo Clinic
treats people from around the world?” These questions cov-
ered all 23 brand associations contained on the laminated
cards used in the BCM procedure. Respondents were asked
to agree or disagree with each statement on a 1 (“strongly
disagree”) to 7 (“strongly agree”) scale. After completing
these ratings, participants completed the same demographic
and background questions as in Study 1. Respondents were
thanked and dismissed.
Results
To assess convergent validity, we compared brand maps
that the respondents in the BCM condition produced with
the rating-scales data obtained in the survey condition.
First, we compared perceptions for the set of brand associa-
tions included in the BCM and rating-scales conditions. We
correlated the frequency of mention of each brand associa-
tion across the brand maps that the BCM respondents con-
structed with the corresponding mean scale rating of those
associations by survey participants. The resulting correla-
tion of .844 (p < .01, N = 23) indicated that the brand asso-
ciations that consumers deemed to be most important in
building their individual brand maps tended to be the same
as those that the survey participants rated highly. For exam-
ple, the association most frequently mentioned on individ-
ual brand maps (“has advanced medical research”) was also
one of the most highly rated associations (M = 6.40) in the
survey.
Second, we extended this basic analysis by computing
weighted frequencies of mention for brand associations
included on individual brand maps in the BCM condition. A
comparison of these weighted frequencies with rating-
scales data enabled us to assess the validity of several fea-
tures of the mapping procedure: (1) the hierarchical place-
ment of brand associations on the consensus map as direct
connections to the brand (Level 1) or connections to other
associations (Levels 2, 3, and 4) and (2) the strength of
brand association links as indicated by the presence of
single, double, or triple lines in the consensus map.
To address the first issue, eachtime a brand association
was included on a map, we weighted it by the level at which
it was placed; higher weights were attached to associations
that were linked more closely to the brand. Weights ranged
from four (directly linked to the brand) to three (linked one
level below in the hierarchy) to two (linked two levels
below in the hierarchy) to one (linked even lower in the
hierarchy). This procedure yielded a weighted frequency of
mention for each brand association, which we then corre-
lated with the corresponding mean scale rating. The result-
ing correlation of .837 (p < .01, N = 23) shows that the hier-
archical placement of brand associations on brand maps
converges well with ratings of the same brand associations
from survey data.
To address the second issue, each time a brand associa-
tion was included on a map, we weighted it by the number
of lines connecting it to the brand or to the association
directly above it. Weights ranged from three (triple line) to
two (double line) to one (single line). This procedure
resulted in a weighted frequency for each brand association,
which we then correlated with the mean scale ratings as we
did previously, producing a correlation of .845 (p < .01, N =
23). Thus, it appears that the selection of connecting lines,
which was meant to denote the strength of the association,
also converges well with the evaluations of rating-scales
respondents.
Discussion
Our results provide evidence of convergent validity for
the BCM. Although the BCM and attribute rating scales are
562 JOURNAL OF MARKETING RESEARCH, NOVEMBER 2006
different in orientation, they agreed on important aspects of
the way consumers view the Mayo Clinic brand. Compari-
sons between these methods add to the validity analyses we
presented in Study 1, providing additional confidence that
the elicitation and mapping procedures measure brand per-
ceptions as intended.
GENERAL DISCUSSION
Contributions to Brand Measurement
The BCM method offers a new option for consumer map-
ping techniques. It delivers a consensus brand map, which
identifies the most important (core) associations that con-
sumers connect to the brand and how these associations are
interconnected. Unlike methods such as ZMET, our
approach gathers consumer perceptions using structured
elicitation, mapping, and aggregation procedures. Standard-
ization offers several advantages. First, the elicitation stage
can use existing consumer research, enabling a firm to
reduce time and expense. Second, because the mapping
stage is structured, respondents can complete the task
quickly (15–20 minutes), without the need for extensive
interviews or specialized interviewing teams. This feature
makes the BCM suitable for different data collection ven-
ues, such as mall intercepts and focus groups, and allows
for the collection of much larger and broader samples.
Finally, because the aggregation process involves the rela-
tively straightforward use of decision rules, obtaining a con-
sensus brand map is less time consuming and less subjec-
tive and does not require specialized statistical training.
These advantages allow for the construction of consensus
brand maps for different market segments, geographic seg-
ments, or constituencies.
The BCM method can also be combined with other
brand-mapping techniques. Consumer mapping techniques,
such as ZMET, offer an unstructured format for eliciting
brand associations, allowing consumers complete freedom
to express their conscious and nonconscious brand percep-
tions in many different ways. In situations in which these
features are desirable, ZMET could be used for developing
a set of brand associations, and the BCM could then be used
to structure the mapping and aggregation stages, providing
a more efficient way to develop a consensus brand map.
Similarly, the BCM could be used for the elicitation and
mapping stages to produce individual brand maps; analyti-
cal mapping techniques, such as network analysis, could
then be used as a more sophisticated approach to producing
a consensus map.
Finally, the BCM is unique among mapping techniques
insofar as it has been evaluated according to traditional tests
for reliability and validity. Standard measurement criteria,
such as convergent and nomological validity, are as impor-
tant for brand-mapping techniques as they are for multi-
item scales, providing assurance that our methods measure
what they are intended to measure.
Contributions to Brand Management
The BCM method offers a picture of how consumers
think about brands, with a visual format that makes it easy
for managers to see important brand associations and how
they are connected in the consumer’s mind. In particular,
one of the most important features highlighted in brand
maps is the core brand associations, the most important set
of brand associations that drive the brand’s image. Although
consumers may identify many things with a brand, it is the
core brand associations, especially those linked directly to
the brand, that should be the focus of management efforts to
build, leverage, and protect brands.
Consider the patient map for the Mayo Clinic (Figure 3).
There are six associations directly connected to the Mayo
Clinic brand. To build or maintain the brand’s image among
patients, management would need to ensure that these asso-
ciations and any associations connected to them continue to
resonate with consumers. For example, to maintain the per-
ception that the Mayo Clinic has the “best doctors in the
world,” branding efforts could be aimed at making this
association salient in communications. In addition, commu-
nications could stress that “doctors work as a team” and that
the Mayo Clinic has “approachable, friendly doctors,”
because these associations are linked with “best doctors in
the world.”
Of equal importance, the core brand associations should
be protected from erosion or dilution. For example, to pro-
tect an association such as “leader in medical research”
from eroding, the organization needs to affirm its commit-
ment to medical research through funding, staff, and public-
ity. An important way that the Mayo Clinic could accom-
plish this would be to continue to commit to being the
“leader in cancer research” and to continue to “publish
health information.” Activities that are incongruent with the
core brand associations need to be questioned for the possi-
bility of diluting important brand associations or adding
new brand associations that are inconsistent with the image.
For example, if the Mayo Clinic opened cosmetic skin care
salons, this would certainly be inconsistent with existing
associations, such as “world leader in medical treatments”
and “expert in treating serious illnesses.”
Changes in the brand over time should be monitored with
respect to the core brand associations uncovered by the
BCM. Surveys that track brand perceptions should assess
consumer perceptions of the core brand associations found
in the consensus brand maps. The BCM methodology can
be repeated on a long-term basis to evaluate whether con-
sumer perceptions of the brand have changed as a result of
branding programs or competitive activity. For example, the
BCM could be used to evaluate the brand’s image every
three to five years, with consumer surveys tracking inter-
mediate changes in core brand associations at 6–12 month
intervals.
Future Research Directions
Several issues remain in refining the BCM methodology
and assessing its suitability for a wide range of branding
contexts. First, it would be useful to evaluate how well the
BCM operates for different types of brands. The Mayo
Clinic has many brand associations that are attribute related
(e.g., “best doctors in the world”), whereas other brands
may have more product-related or experience-related asso-
ciations. We have applied the BCM to several brands,
including Nike, Disney, and Sony, with promising results.
For example, with Nike, we carried out the elicitation and
mapping procedures with college students, who participated
ina class setting. We used the same aggregation procedures
as those described in the Mayo Clinic application, produc-
Brand Concept Maps 563
3For example, analyses of split-half reliability yielded similar results to
those reported for the Mayo Clinic. Correlations computed across split-half
maps for Nike were highest for the presence of core brand associations
(φ = .84, p < .01; N = 30) and presence of first-order brand associations
(φ = .80, p < .01; N = 30) and moderate for the presence of specific brand
association links (φ = .49, p < .01; N = 435).
ing a consensus brand map for Nike with acceptable levels
of reliability and validity.3
Second, it would be useful to incorporate procedures into
the BCM to assess the nature of relationships between asso-
ciations, that is, whether it is causal, correlational, or some-
thing else. Although we can speculate about the relation-
ships shown in the consensus brand maps, we have not yet
developed a technique for doing so on an objective basis.
For example, it seems clear that perceptions of Mayo Clinic
as “treats famous people around the world” cause people to
believe that Mayo Clinic is “known worldwide.” However,
being a “leader in cancer research” could be an instance of
being a “leader in medical research,” or one of these associ-
ations could be driving (causing) the other. We believe that
procedures similar to those used in understanding causal
reasoning chains (see Sirsi, Ward, and Reingen 1996) could
be incorporated into the mapping stage of the BCM to pro-
vide information about brand association relationships.
Third, modifications of the BCM mapping procedure
could be developed to make data collection even easier and
more flexible. In the Nike research, we modified the map-
ping procedure to be amenable to data collection in a large
group setting (i.e., list of brand associations were shown on
a projection screen). Further development along these lines,
especially with computer-aided data collection, would be
valuable as well.
Although work in these areas remains to be done, we
believe that the BCM methodology holds promise and is
worthy of further research to understand its uses and limita-
tions better. We look forward to meeting these challenges.
REFERENCES
Aaker, David A. (1996), Building Strong Brands. New York: The
Free Press.
Anderson, John R. (1983), The Architecture of Cognition. Cam-
bridge, MA: Harvard University Press.
Christensen, Glenn L. and Jerry C. Olson (2002), “Using the Zalt-
man Metaphor Elicitation Technique (ZMET) to Map Con-
sumers’ Mental Models,” Psychology & Marketing, 19 (June),
477–502.
Churchill, Gilbert A., Jr. (1979), “A Paradigm for Developing Bet-
ter Measures of Marketing Constructs,” Journal of Marketing
Research, 16 (February), 64–73.
Eagly, Alice H. and Shelly Chaiken (1993), The Psychology of
Attitudes. Orlando, FL: Harcourt Brace Jovanovich.
Fishbein, Martin and Icek Ajzen (1975), Belief, Attitude, Intention,
and Behavior. Reading, MA: Addison-Wesley.
Henderson, Geraldine R., Dawn Iacobucci, and Bobby J. Calder
(1998), “Brand Diagnostics: Mapping Branding Effect Using
Consumer Associative Networks,” European Journal of Opera-
tional Research, 111 (December), 306–327.
Joiner, Christopher (1998), “Concept Mapping in Marketing: A
Research Tool for Uncovering Consumers’ Knowledge Struc-
ture Associations,” in Advances in Consumer Research, Vol. 25,
Joseph W. Alba and Wesley Hutchinson, eds. Ann Arbor, MI:
Association for Consumer Research, 311–17.
Keller, Kevin L. (1993), “Conceptualizing, Measuring, and Man-
aging Customer-Based Brand Equity,” Journal of Marketing, 57
(January), 1–22.
Novak, Joseph D. and D. Bob Gowin (1984), Learning How to
Learn. New York: Cambridge University Press.
Reynolds, Thomas J. and Jonathan Gutman (1988), “Laddering
Theory, Method, Analysis, and Interpretation,” Journal of
Advertising Research, 28 (February–March), 11–31.
Rokeach, Milton (1968), Beliefs, Attitudes, and Values: A Theory
of Organization and Change. San Francisco: Jossey-Bass.
Ruiz-Primo, Maria Araceli and Richard J. Shavelson (1996),
“Problems and Issues in the Use of Concept Maps in Science
Assessment,” Journal of Research in Science Teaching, 33
(August), 569–600.
Sirsi, Ajay K., James C. Ward, and Peter H. Reingen (1996),
“Microcultural Analysis of Variation in Sharing of Causal Rea-
soning About Behavior,” Journal of Consumer Research, 22
(March), 345–72.
Wallace, Josephine D. and Joel J. Mintzes (1990), “The Concept
Map as a Research Tool: Exploring Conceptual Change in Biol-
ogy,” Journal of Research in Science Teaching, 27 (December),
1033–1052.
Zaltman, Gerald (1997), “Rethinking Market Research: Putting
People Back In,” Journal of Marketing Research, 35 (Novem-
ber), 424–37.
——— and Robin Higie Coulter (1995), “Seeing the Voice of the
Customer: Metaphor-Based Advertising Research,” Journal of
Advertising Research, 35 (July–August), 35–51.